Rows are charged by the symbol, so an empty class decoded its collective bitmap for nothing: 65,535
rows a class, a thousand classes a dictionary. 806 KB took 196 s. It steps over the bytes instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AmsdeUCdTaD72jkDw2pJKY
From a review of the work count. An item with no pixels was charged a step, though reading it can take
hundreds of decisions, so a 22,515-byte PDF of refined text instances of no size stayed inside the
limit and took 730 s to render. Empty height classes and export runs weren't charged at all, and 19 KB
of them took 47 s. Each integer an item reads now costs what one takes at worst, 38, so a decision
never costs more than a pixel. A full region of 10,000 instances is 0.14% of the limit, so a real file
doesn't notice.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AmsdeUCdTaD72jkDw2pJKY
From CodeRabbit on the work count.
A symbol with no rows cost no work, though it is decoded all the same, so a stream of thousands of
dictionaries of such symbols, exporting none, went uncounted. Each symbol now costs a step more than
its pixels.
A refined text instance was charged after it was decoded, so one could still be 2^28 pixels past what
was left before the decoder refused. It's charged once its size is read, before anything is allocated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AmsdeUCdTaD72jkDw2pJKY
Each JBIG2 bitmap is bounded, but nothing bounded how many a stream asks for, so a few hundred
bytes could tie up a render worker for minutes, and CodeRabbit raised the same on #5461. Measured
through pdfcraft-cli, before and after:
- one halftone segment, a 795-byte PDF: 265 s, now refused at once;
- 66 halftones over patterns 1 wide, a 3,894-byte PDF: about 330 s, now refused at once;
- about 226,000 generic regions behind nested Flate, a 923-byte PDF: unfinished at 40 s, some
88 hours by extrapolation, now 4.8 s;
- 8 dictionaries of symbols of no width, a 983-byte PDF: 81 s, now 0.7 s;
- about 1,048,000 generic regions of no width, a 34 KB PDF: unfinished at 90 s, now 2.6 s;
- an aggregated dictionary over 2^20 symbols, a 713-byte PDF: unfinished at 90 s, now 9.1 s;
- 10,000 instances of an 8192 × 8192 symbol, an 18 KB PDF: 64 s, now 3.2 s;
- 30 generic regions of no data: 48 s, now 4.7 s.
A stream may now ask for 2^30 units of work: four of the largest pages, about 6 s on data the
decoder predicts and about 17 s on data it can't, inside the viewer's 20 s watchdog. Decoding a
bitmap costs a unit a pixel and 4 a row, since a row costs time at any width. Copying one costs a
unit a word and 1 a row, so a tall thin bitmap costs its rows. Each item is charged before it is
decoded, apart from what is known only as it goes: new symbols and text instances are charged in
their loops. An aggregated symbol is charged for the symbol table it rebuilds. Past the bound the
image isn't drawn, as with the other caps, and the rest of the page is.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AmsdeUCdTaD72jkDw2pJKY
Each JBIG2 bitmap was bounded on its own, but a few kilobytes could still exhaust memory:
nothing bounded what a page's segments hold for later ones, or how far one segment could
multiply them. Measured on black-box streams, peak memory before and after:
- intermediate regions, 16 in 774 bytes: 514 MiB, now 98 MiB;
- an 8 MiB symbol exported 1,200 times, a 2,989-byte PDF: killed at 8 GiB, now 79 MB;
- dictionaries of zero-width symbols: 781 MB, now 72 MB;
- dictionaries retaining their coding contexts, 1.1 MB: 2.89 GB, now 71 MB;
- 65,535 references to one dictionary, 78 KB: about 17 GB, now 6 MB;
- a halftone grid of 16,384 cells a side, 98 bytes: 1.1 GB, now 2.4 MB;
- a pattern dictionary of 65,535 patterns, 48 bytes: 539 MB, now 15 MB, and it still draws;
- a custom Huffman table behind nested Flate, a 1,297-byte PDF: killed at 8 GiB, now 278 MB,
which is mostly the decoded stream;
- 11-byte segments behind nested Flate in a stream and its globals, a 4,765-byte PDF: 5 GB, now
602 MB, which is mostly the two decoded streams;
- an Indexed image whose page header claims 2^20 a side, a 743-byte PDF: aborted asking for
137 GB, now 23 MB;
- a chain of 4,000 /JBIG2Globals streams that are JBIG2 themselves, on a 2 MiB stack: overflowed
it and aborted, now 24 MB.
So PageState counts what it stores: regions, pattern and symbol dictionaries and Huffman
tables, with bitmap headers, retained contexts and a charge per item, up to 64 MiB, two square
bitmaps of 16,384 a side. A dictionary's new and exported symbols count against what is left as they are
made, not once it is stored. A segment may refer to at most 2^20 symbols, checked before any are
collected. A halftone grid has at most 2^25 cells. A pattern dictionary keeps at most 16 MiB of
shifted copies, which only speed up drawing, and none that wouldn't fit what the page has left. A custom Huffman table has at most 2^16 lines, and a
stream with its globals at most 2^20 segments and 2^22 references, checked as they are read. A
page past the bounds on a bitmap is refused when the stream is opened, since callers size their
output from its header before the page bitmap is made. And in hayro-syntax, a /JBIG2Globals
stream that is itself JBIG2 isn't decoded: globals are raw segments, so it reads as globals that
can't be decoded.
An image past a bound isn't drawn, as with the other caps, and the rest of the page is. This
bounds memory, not time: decoding work has no total yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AmsdeUCdTaD72jkDw2pJKY
Includes #4325#1952#4180#2529#4320#4805, preserving their original histories.
Keep paragraph alignment and hard/soft wrap metadata together; retain CCITT/downsampling/RGB ownership guards with bounded JBIG2 export and exact-copy symbol encoding.
Combined interaction regressions added. Validation pending.
Use a named compact preset arrow beside the field, retain the row alignment/width assertions for the editable control, and open the current colour flyout in the saved-text regression. No accessibility or layout checks are disabled.
Preserve caption editing and current form/Prepare module boundaries. Carry export-only appearance preservation, validation and data round-trip tests to their current paths, and retain current attribution metadata. Hosted CI must validate this refreshed head.
Preserve main's continuous text selection, page tools and module split. Carry only this branch's sidebar-focus feature and regression onto main, without resurrecting old common files from criss-cross merge bases. Hosted CI remains required.
Use the unified Format Text panel and current page-input context instead of restoring old monolithic UI code. Preserve main's pointer routing, editor accessibility and preferences; retain colour, save and input regressions. Hosted CI must validate this refreshed head.
UPDATE_APP (native builds; craft-update is a native-only dependency) carries our 128 px icon and accent (#b70228, its red, from the icon (#5245)). The updater, the self-update
helper and craft-update's startup prompt all get them through crate::UPDATE_APP; the
prompt draws the icon, the accent and "by ArtCraft" (craft-update's default developer).
craft-update is pinned to storytold/craft-libs main at 1065653e7 (#434).
Clippy fixes for Rust 1.99 in untouched code (same logic):
- crates/edit/src/images.rs
- crates/forms/src/appearance.rs
- `zip`: 9.1.0 → 9.1.3
Requirement `roxmltree` 0.20 → 0.21: `ParsingOptions` gained `entity_resolver` (initializers take `..Default::default()`, i.e. none, as before) and is no longer `Copy` (two functions build it with a closure, the pattern merge.rs already used).
Packing makes an RGB image grey for every use at once, but /DefaultGray and
/DefaultRGB were checked only where the walk reads: page content and forms.
An annotation's appearance, a pattern, a Type3 glyph or a soft mask's group
drawing the same image can have other default colour spaces, so its colours
could change there.
The walk now records, by object number, every reference in the resources it
reads (each /XObject entry, drawn or not, an /XObject dictionary under its
resources, resources under a page, page-tree node or form), reading the
resources of every page and of every node above one, each page and node a
root. An image the walk reached stays covered unless an object reachable
from the trailer refers to it, or to an object on its way from a page, other
than through those references, or it is held by one that isn't covered; an
eight-bit DeviceRGB image that isn't covered stays RGB. So resources, an
/XObject dictionary or a form a page shares with an appearance keep it RGB,
as does a reference with another generation, while LibreOffice's layout (one
resources object for a page, its /Pages node, and a form inside it that
names it again) still packs. The scan runs only when packing is on and there
is such an image.
The images drawn or listed where the defaults are in force are kept by
object number too, so one named with another generation there stays RGB.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The optimizer now writes JBIG2 images as a symbol dictionary and text
regions, and Export ▸ Images has to read them back. A page of three
repeated letters, symbol coded (smaller than the generic coding),
exports as its picture. The test is added to create.export-all-images'
parity evidence. Asked for in pre-push review.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
Review of the previous commit:
- Nothing checked the segment numbers the region split assigns: a
mutant numbering every region 2 decoded. The many-marks test now
reads each segment's number, and has a bar too tall to be a symbol,
so the generic region is numbered after the last text region.
- The decoder's 32,767-symbol limit was checked against all distinct
marks, before either plan, so an image with more distinct marks lost
the second plan, which codes only the repeated ones. The limit now
applies to each plan (MAX_CODED_SYMBOLS); finding marks keeps its
memory bound of 65,536 (MAX_SYMBOLS), as before the previous commit.
A test has 32,768 distinct marks above a page of repeated text, which
encode_jbig2 still symbol codes through the second plan.
- The dense-page optimizer test now checks that it takes several text
regions, and a compile-time assert keeps segment numbers within a
byte's reach.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
A review found that a dense page came out as a stream PdfCraft's own
renderer refuses, so the page drew blank after optimizing: hayro-jbig2
decodes a text region of at most 10,000 symbol instances, and refuses
32,768 symbols or more. The stream was valid (Poppler decoded it), but
every mark went into one text region and up to 65,536 symbols were kept.
A 300 dpi letter page of the optimizer test's own text crossed it.
A page with more than 10,000 marks is now placed by several text
regions, each over a band of rows and all referring to the one symbol
dictionary, which each region but the last keeps retained (7.2.4). Past
32,767 distinct symbols the image is coded generically.
New tests: 25,000 copies placed by three regions, which decode; the
symbol caps decoded, not just reached; and in the optimizer, a whole
letter page of rendered text that looks the same as JBIG2 symbols (it
drew blank before). The READMEs say "still looked at" for the caps, and
that the scans figure is the six pages' total (20% on one page).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
A second review found two rules tested one way only: a mark 257 tall
(not just wide) is left to the generic region, which also bounds the
bitmap allocated for each mark, and runs joined at the lower-left corner
are one mark. The test now has a 1 x 257 column and an anti-diagonal.
The caps on distinct symbols (65,536) and their pixels (2^24) are tested
at and past them, as the run and row caps are.
The scans figure now names its noise: with 2% or more of the letters'
edge pixels flipped, symbols save at most 8% over the generic region.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
Review found behaviour the tests let break unnoticed (each was a
surviving mutant, and each new test fails on it):
- black padding bits: a run reaching past the row would stop symbol
coding on many scans;
- Annex A's 32-bit range and PREV at exactly 256, where a decision would
go uncoded: coordinates past 4,435 on a 12,000-pixel-wide and a
10,050-row page, and the contexts of 4436's coding;
- marks joined at a corner are one mark; 256 pixels a side is a symbol,
257 is left to the generic region;
- the run and row caps, at and past them.
The roadmap's status line for this change now comes first, and the
"about a third smaller than Group 4 on scans" figure is replaced by what
was measured: 27-33% on the same pages with scan-like noise added.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
pdfcraft_filters::encode_jbig2 stores each distinct mark of a 1-bit
image (8-connected black pixels, at most 256 a side) once, as a symbol
in a symbol dictionary, and places its exact copies with an immediate
lossless text region that refers to it. Larger marks, and in a second
plan marks that occur once, go to a generic region over their box. It
keeps whichever of that and the generic coding (encode_jbig2_generic,
unchanged) is smaller. Only exact copies share a symbol, so the stream
decodes to the same bits.
Written from the JBIG committee's ISO/IEC 14492 Final Committee Draft:
6.4 and 6.5 (text region and symbol dictionary decoding), 7.2, 7.4.2
and 7.4.3 (segment headers and syntax) and Annex A (the integer coders,
checked against its worked example and Table A.1).
On six pages of rendered 300 dpi text the output is 28% of the generic
coding's size and 17% of Group 4's, at about 100 ms a page; hayro-jbig2
and Poppler (pdfimages) both decode it to the original bits. Scans,
whose copies of a letter are seldom identical, gain little: that needs
refinement coding. A dictionary shared between pages (/JBIG2Globals)
is a further step.
The optimizer's Monochrome Images > JBIG2 (lossless) and doc_optimize
mono jbig2 use it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
An image's /JBIG2Globals were decoded with the 1 GiB any stream may decode
to, before the JBIG2 page's size is known, so the page's pixel limit didn't
bound them: a few kilobytes of Flate could make export allocate up to 1 GiB.
They now decode within MAX_GLOBALS (32 MiB); globals past it skip the image,
with the limit in the reason. The test covers one byte over and exactly the
limit, and the image's /DecodeParms direct, by reference and in an array.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The test only checked each refusal's message. Without the tool's notation guard, set_props
refuses the same notation with the same message as a failed edit, so the test passed with the
guard removed. It now asserts InvalidArgs for every row, and fails on "a EVCALC b" without the
guard.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ckb and nb get all eleven rows, th and lo the six strings, and sq the five plurals beside its
strings, each under "# Untranslated". Their rows next to these were missing too, so following
them had added nothing.
hi, id, ko, lo and th list one plural form (plural_none), so they get no English copy of the
plurals: one English wording would be used for every count ("1 formulas read ..."). Without an
entry, the English fallback picks the right form, as it already does for the page's other plurals
in those catalogues.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
set_props refuses such a notation before writing anything, so the
Calculate tab can't store one either (the tool already refused it). A
refused edit used to close Field Properties and lose everything typed
into it; the dialog now stays open with its changes, as the dialog for
several fields does, and the notice says why (from the review).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
parse_calculate trimmed the notation with str::trim, which also took an
escaped space ending its last name: "Qty * Total\ " read back as
"Qty * Total\", a field the form doesn't have, while the script's
JavaScript still read "Total ". It now trims to its first and last
token, as the field-reading scan does. notation_storable compares the
names and operators read back with those given, so a notation ending in
a lone \ (which escapes the space after it) is refused; a proptest reads
back notations of any names (from the review).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
A notation is kept in the script's /** BVCALC … EVCALC **/ comment and read
back from there. One holding EVCALC was stored cut short ("a EVCALC b"
read back as "a", and its JavaScript translation didn't parse), one
holding */ ended the comment, and an empty one became "event.value = ;".
form_set_props now refuses each (af::notation_storable: it has to read back
as itself). The test also pins a notation beside op alone and beside
fields alone (from the review).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
steps() takes the widgets by object, so each of two fields sharing a name
gets its own step; a test checks it, unfocused and with a focus (from the
review: no test failed with the step looked up by name).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
A notation beside op and fields set the notation and dropped the rest, a
field name that wasn't a string was left out, a misspelt key was ignored,
and validate {"min": 0, "max": "100"} wrote a min-only range. Each is
now refused and nothing changes. A null bound, as form_fields shows the
one a range doesn't have, is no bound, so its output is taken back.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
AFSimple_Calculate("SUM", "a, b", x) gave only a built-name site, so a
rename no longer rewrote a and b (main did) while parse_calculate still
read the sum of a and b: after a rename the sum read missing fields. The
string list now keeps its names when an argument follows, and what
follows is a name built at run time, as for an array (from the review).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
The page's focus was a name, so with two fields called total (a
malformed form) hovering the second one's row showed the first one's
missing fields in the pill, and outlined and numbered both on the page.
A row and a widget under the pointer now carry their field object: the
pill says that field's reads and issues, and the page outlines it, its
step and the formulas that read it by object. The Prepare selection is
still only a name, so it stands for every field of that name and the
pill says all of their issues (from FredApps's review).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P6yUtvWGPsJpMLyh6pxSXG
found accepted price.0 (up to the field's last widget), but links,
recalculate and the scripts' getField all take a name as the field or
the fields under it, so a formula reading price.0 got no value and no
order check while the page said nothing was wrong. found now looks
names up the same way; a widget name is flagged as missing until all of
them resolve it (from CodeRabbit's review).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A read like price.12 counted as found whenever price existed. A widget
number now finds something only below the field's widget count; a
field with one widget, or none, still answers to .0 (from CodeRabbit's
review).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
calculate_json wrote the lowercased AFSimple code, so a product or an
average came back as {"op": "prd"} or {"op": "avg"}, which
form_set_props refused as an unknown op. One name per operation now
serves both: calculate_arg looks names up in it (still taking minimum
and maximum), and the listings write it (from CodeRabbit's review).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
calculate_arg read the notation first, so {"notation": …, "script": …}
set the notation and dropped the script unnoticed; the script check now
comes first (from CodeRabbit's review).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BfMGHQNgdq5yf5tKZHGR3z
